1. optimize the error message of softmax_with_cross_entropy_op;2. add input...

1. optimize the error message of softmax_with_cross_entropy_op;2. add input value check for cross_entropy, if the dimention of input is zero, raise error. test = develop

paddle/fluid/operators/softmax_with_cross_entropy_op.h

@@ -61,7 +61,7 @@ class SoftmaxWithCrossEntropyKernel : public framework::OpKernel<T> {
       PADDLE_ENFORCE_GT(
           n, 0, platform::errors::InvalidArgument(
                     "The size of axis should be larger than 0, but received "
-                    "axis size is %d.",
+                    "SizeToAxis of softmax is %d.",
                     n));
 
       const int d = SizeFromAxis(axis, softmax->dims());
@@ -110,7 +110,7 @@ class SoftmaxWithCrossEntropyKernel : public framework::OpKernel<T> {
     PADDLE_ENFORCE_GT(
         n, 0, platform::errors::InvalidArgument(
                   "The size of axis should be larger than 0, but received "
-                  "axis size is %d.",
+                  "SizeToAxis of logits is %d.",
                   n));
 
     const int d = SizeFromAxis(axis, logits->dims());
@@ -162,7 +162,7 @@ class SoftmaxWithCrossEntropyGradKernel : public framework::OpKernel<T> {
     PADDLE_ENFORCE_GT(
         n, 0, platform::errors::InvalidArgument(
                   "The size of axis should be larger than 0, but received "
-                  "axis size is %d.",
+                  "SizeToAxis of logit_grad is %d.",
                   n));
 
     const int d = SizeFromAxis(axis, logit_grad->dims());

python/paddle/nn/functional/loss.py

@@ -1644,6 +1644,9 @@ def cross_entropy(input,
             ignore_index)
 
     input_dims = len(list(input.shape))
+    if input_dims == 0:
+        raise ValueError('The dimention of input should be larger than zero!')
+
     label_dims = len(list(label.shape))
     if input_dims - 1 != label_dims and input_dims != label_dims:
         raise ValueError(
@@ -2016,3 +2019,20 @@ def sigmoid_focal_loss(logit,
         loss = paddle.sum(loss, name=name)
 
     return loss
+
+
+if __name__ == "__main__":
+    input_arr = np.array([], dtype=np.float32)
+    input = paddle.to_tensor(np.reshape(input_arr, (0, 0)), dtype='float32')
+
+    label = paddle.to_tensor([], dtype='float32')
+
+    weight = paddle.to_tensor([], dtype='float32')
+
+    result = cross_entropy(
+        input,
+        label,
+        weight=weight,
+        ignore_index=-100,
+        soft_label=False,
+        axis=-1)